A freely available tool to document wartime destruction
PNAS Nexus
image:
Destruction analysis of all 10m×10m building pixels in Beirut over 12-day periods from July 11 to July 23 (left), July 23 to August 4 (middle), August 4 to August 16 (right; all 2020). Lower p-values indicate a higher likelihood that part of a building was destroyed. The harbor explosion on August 4 is denoted by the red dot in the middle image, with radii of the blast wave with varying distances (also in red). Buildings located directly next to the sea are missing some pixels due to the processing of the images.
view moreCredit: Racek et al.
Researchers develop a method to detect the destruction of buildings using freely available satellite radar imagery. Daniel Racek and colleagues’ algorithm analyzes publicly available Sentinel-1 synthetic aperture radar images from the European Space Agency to identify destroyed buildings in conflict zones. The method statistically assesses the visual similarity of locations over time, enabling detection of destruction from a single satellite image every 12 days, without requiring labeled training data or expensive proprietary imagery. Unlike optical satellites, radar operates through clouds and darkness. The authors validate the approach across three case studies: the 2020 Beirut harbor explosion, the 2022 siege of Mariupol, Ukraine, and the 2023–2024 Gaza conflict. In Beirut, the algorithm achieved precision of 86%, correctly identifying most buildings destroyed by the explosion. In Mariupol's Zhovtnevyi district, the method estimated 2,437 buildings were destroyed, some 22% of all buildings in the district. In Gaza, destruction estimates tracked closely with UN satellite analysis. According to the authors, the method democratizes access to conflict monitoring tools and enables near real-time assessment of building destruction for humanitarian response, human rights monitoring, and academic research on armed conflict.
Destruction analysis of all 10m×10m building pixels in Gaza over 12-day periods from September 18, to December 11, 2023. Lower p-values indicate a higher likelihood that part of a building was destroyed. The timeline at the bottom denotes key events taking place between image acquisition dates.
Credit
Racek et al.
Journal
PNAS Nexus
Article Title
Unsupervised detection of building destruction during war from publicly available radar satellite imagery
Article Publication Date
9-Dec-2025
No comments:
Post a Comment